2 * Copyright © 2011 Google, Inc.
4 * This is part of HarfBuzz, a text shaping library.
6 * Permission is hereby granted, without written agreement and without
7 * license or royalty fees, to use, copy, modify, and distribute this
8 * software and its documentation for any purpose, provided that the
9 * above copyright notice and the following two paragraphs appear in
10 * all copies of this software.
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
24 * Google Author(s): Behdad Esfahbod
27 #include "hb-ot-shape-private.hh"
28 #include "hb-ot-shape-complex-private.hh"
35 * This file exports one main function: _hb_ot_shape_normalize().
37 * This function closely reflects the Unicode Normalization Algorithm,
38 * yet it's different. The shaper an either prefer decomposed (NFD) or
41 * In general what happens is that: each grapheme is decomposed in a chain
42 * of 1:2 decompositions, marks reordered, and then recomposed if desires,
43 * so far it's like Unicode Normalization. However, the decomposition and
44 * recomposition only happens if the font supports the resulting characters.
48 * - Try to render all canonically equivalent strings similarly. To really
49 * achieve this we have to always do the full decomposition and then
50 * selectively recompose from there. It's kinda too expensive though, so
51 * we skip some cases. For example, if composed is desired, we simply
52 * don't touch 1-character clusters that are supported by the font, even
53 * though their NFC may be different.
55 * - When a font has a precomposed character for a sequence but the 'ccmp'
56 * feature in the font is not adequate, form use the precomposed character
57 * which typically has better mark positioning.
59 * - When a font does not support a character but supports its decomposition,
60 * well, use the decomposition.
62 * - The Indic shaper requests decomposed output. This will handle splitting
63 * matra for the Indic shaper.
67 output_glyph (hb_ot_shape_context_t *c,
70 hb_buffer_t *buffer = c->buffer;
72 buffer->output_glyph (glyph);
73 hb_glyph_info_set_unicode_props (&buffer->out_info[buffer->out_len - 1], buffer->unicode);
77 decompose (hb_ot_shape_context_t *c,
81 hb_codepoint_t a, b, glyph;
83 if (!hb_unicode_decompose (c->buffer->unicode, ab, &a, &b) ||
84 (b && !hb_font_get_glyph (c->font, b, 0, &glyph)))
87 bool has_a = hb_font_get_glyph (c->font, a, 0, &glyph);
88 if (shortest && has_a) {
96 if (decompose (c, shortest, a)) {
113 decompose_current_glyph (hb_ot_shape_context_t *c,
116 if (decompose (c, shortest, c->buffer->info[c->buffer->idx].codepoint))
117 c->buffer->skip_glyph ();
119 c->buffer->next_glyph ();
123 decompose_single_char_cluster (hb_ot_shape_context_t *c,
126 hb_codepoint_t glyph;
128 /* If recomposing and font supports this, we're good to go */
129 if (will_recompose && hb_font_get_glyph (c->font, c->buffer->info[c->buffer->idx].codepoint, 0, &glyph)) {
130 c->buffer->next_glyph ();
134 decompose_current_glyph (c, will_recompose);
138 decompose_multi_char_cluster (hb_ot_shape_context_t *c,
141 /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
142 for (unsigned int i = c->buffer->idx; i < end; i++)
143 if (unlikely (is_variation_selector (c->buffer->info[i].codepoint)))
146 while (c->buffer->idx < end)
147 decompose_current_glyph (c, FALSE);
151 compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
153 unsigned int a = pa->combining_class();
154 unsigned int b = pb->combining_class();
156 return a < b ? -1 : a == b ? 0 : +1;
160 _hb_ot_shape_normalize (hb_ot_shape_context_t *c)
162 hb_buffer_t *buffer = c->buffer;
163 bool recompose = !hb_ot_shape_complex_prefer_decomposed (c->plan->shaper);
164 bool has_multichar_clusters = FALSE;
167 /* We do a fairly straightforward yet custom normalization process in three
168 * separate rounds: decompose, reorder, recompose (if desired). Currently
169 * this makes two buffer swaps. We can make it faster by moving the last
170 * two rounds into the inner loop for the first round, but it's more readable
174 /* First round, decompose */
176 buffer->clear_output ();
178 for (buffer->idx = 0; buffer->idx < count;)
181 for (end = buffer->idx + 1; end < count; end++)
182 if (buffer->info[buffer->idx].cluster != buffer->info[end].cluster)
185 if (buffer->idx + 1 == end)
186 decompose_single_char_cluster (c, recompose);
188 decompose_multi_char_cluster (c, end);
189 has_multichar_clusters = TRUE;
192 buffer->swap_buffers ();
195 /* Technically speaking, two characters with ccc=0 may combine. But all
196 * those cases are in languages that the indic module handles (which expects
197 * decomposed), or in Hangul jamo, which again, we want decomposed anyway.
198 * So we don't bother combining across cluster boundaries.
200 * TODO: Am I right about Hangul? If I am, we should add a Hangul module
201 * that requests decomposed. */
203 if (!has_multichar_clusters)
207 /* Second round, reorder (inplace) */
210 for (unsigned int i = 0; i < count; i++)
212 if (buffer->info[i].combining_class() == 0)
216 for (end = i + 1; end < count; end++)
217 if (buffer->info[end].combining_class() == 0)
220 /* We are going to do a bubble-sort. Only do this if the
221 * sequence is short. Doing it on long sequences can result
222 * in an O(n^2) DoS. */
228 hb_bubble_sort (buffer->info + i, end - i, compare_combining_class);
237 /* Third round, recompose */
239 /* As noted in the comment earlier, we don't try to combine
240 * ccc=0 chars with their previous Starter. */
242 buffer->clear_output ();
244 unsigned int starter = 0;
245 buffer->next_glyph ();
246 while (buffer->idx < count)
248 if (buffer->info[buffer->idx].combining_class() == 0) {
249 starter = buffer->out_len;
250 buffer->next_glyph ();
254 hb_codepoint_t composed, glyph;
255 if ((buffer->out_info[buffer->out_len - 1].combining_class() >=
256 buffer->info[buffer->idx].combining_class()) ||
257 !hb_unicode_compose (c->buffer->unicode,
258 buffer->out_info[starter].codepoint,
259 buffer->info[buffer->idx].codepoint,
261 !hb_font_get_glyph (c->font, composed, 0, &glyph))
263 /* Blocked, or doesn't compose. */
264 buffer->next_glyph ();
268 /* Composes. Modify starter and carry on. */
269 buffer->out_info[starter].codepoint = composed;
270 hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer->unicode);
272 buffer->skip_glyph ();
274 buffer->swap_buffers ();